The importance of context in store

forecasting: The site visit in retail

location decision-making
Received (in revised form): 5th December, 2007

Steve Wood
is a lecturer at the School of Management, University of Surrey and an Advanced Institute of Management (AIM) Scholar. Before re-entering academia
in 2005, he worked as a location analyst for Tesco Plc.

Andrew Tasker
is Director of Optimum Locations, an independent retail location analysis consultancy. He previously led the Location Planning function at Sainsbury’s
for nearly 15 years. He is a founder member of the Demographics User Group and a member of the Marketing Society and Market Research Society.

Keywords store location, site visit, store forecasting, knowledge management, location planning

Abstract The aim of this paper is to investigate how practical store location decision-making
balances formal modelling with the less well-studied informal qualitative inputs. By using case studies
from one major UK food retailer, we find that informal knowledge has to be considered seriously
alongside quantitative models despite the inclusion of such knowledge often proving challenging. In
particular, the site visit has a key role in contextualising factors that are difficult to represent in formal
‘modelled’ data, and in calibrating the inputs to models that are becoming increasingly advanced. We
conclude that conceptualising the role of knowledge management in retail store decision-making has
been under-theorised but can offer a key to advancing our understanding of this process still further.
Journal of Targeting, Measurement and Analysis for Marketing (2008) 16, 139–155. doi:10.1057/jt.2008.3;
published online 10 March 2008

INTRODUCTION is tied to sales. Therefore, a site evaluation begins

Readers of geographical-, marketing- and service with an estimate of sales that can reasonably be
industry-related journals will be impressed by the expected’ (p. 1).
advances made in the discipline of retail store While degrees of retail investment may vary
location forecasting and decision-making over the markedly between a 2,000 ft2 convenience store
past two decades.1 The increased development of and a large 120,000 ft2 hypermarket, new store
computing power, aligned with ever-increasing real construction represents a huge capital outlay to
estate values (and therefore the huge sunk costs of the operator.4 To underline the critical nature
store development)2 has led to an environment of such decision-making in a contemporary
where leading retailers are increasingly willing to perspective our industry experience suggests that
invest capital in improved accuracy in forecasting, a 10 per cent variation in a sales forecast from
which thereby reduces investment risk. As Cohen reality for a medium-sized grocery superstore
and Applebaum3 described it back in 1960, ‘Profit could change the affordable bid for a site by
about £5m. Given the highly competitive nature
of land sale auctions for strategically located sites,
Correspondence: Steve Wood, School of Management, this could potentially mean losing the site to a
University of Surrey, Guildford, Surrey GU2 7XH, UK. competitor, or conversely, leaving the retailer with
Tel: + 44 (0)1483 686344;
Fax: + 44 (0)1483 686346;
an unprofitable store. Unsurprisingly, a profitable
E-mail: [email protected] niche has emerged for the leading store location

© 2008 Palgrave Macmillan Ltd 0967-3237 $30.00 Vol. 16, 2, 139–155 Journal of Targeting, Measurement and Analysis for Marketing 139
www.palgrave-journals.com/jt
 Wood and Tasker

companies such as MapInfo, Experian and CACI the role of the site visit and the types of
that act in a consulting capacity to retailers, knowledge and findings that it can reveal. In the
having developed extensive expertise in store subsequent fourth and fifth sections, we review
forecasting and assessment techniques.5 how these knowledges are successfully integrated
While recognising the tremendous benefit that (or not) into the formalised forecasting process
highly quantitative, technological and data-rich and the broader challenges that this entails. Finally,
methods can have for ‘in-office’ decision-support, we end with some conclusions and managerial
in this paper we contend that this must not be to implications.
the detriment of thorough and methodological
investigations at the level of the site visit. As
much organisation theory has underlined, there is THE BACKGROUND TO
a difference between theoretical work routine and DECISION-SUPPORT SYSTEM
actual ‘work in practice’.6–8 First, data and insight SOPHISTICATION
gleaned while physically ‘on location’ allows the Hernandez et al.11 distinguish between a number
calibration of data inputs that feed into spatial of different hierarchies of retail location planning
interaction models. Furthermore, site and and decision-making, ranging from ‘strategic’,
catchment examination may permit the which is concerned with the macro scale issues
identification of factors that are not necessarily of a retailer’s locational strategy and typical
revealed in data purchased from consultancies or positioning, and at the other extreme, ‘tactical’,
acquired from the Census. Secondly, insights which focuses on issues of local marketing and
gleaned when on-site may often be difficult to promotion at the micro scale. In the middle of
represent in many of the ‘in office’ highly the scale is the issue of ‘monadic’ decision-
quantitative models, thereby presenting challenges making, concerning individual outlets, whether
in decision-making. This underlines the via new openings, closures or refits. It is this final
continuing conflict within Management Science category with which we are concerned in this
more broadly between tacit knowledge that is paper and the tools that retailers use to aid in
‘deeply rooted in action, commitment, and store forecasting (see Table 1).
involvement in context’ and therefore ‘hard to Established sales forecasting decision-support
formalise and communicate’,9 and codified include simply utilising experience12 and
knowledge ‘that is transmittable in formal, checklists13 on field visits, as well as analogue
systematic language’ (Nonaka,10 p. 16) and thus models.14 As data became easier to collect and
more easily introduced into quantitative sophistication increased, there were increasingly
modelling solutions. Hence, this research aims to complex attempts at multiple regression modelling
unpack the information that emerges from this which saw considerable returns in terms of
critical stage of the assessment process and links increased sales forecasting accuracy.15 At the
with the knowledge management literature extremes, there were also some early efforts at
concerning how such insights can be utilising the theoretical contribution of spatial
incorporated into decision-making processes. interaction modelling, though numerous studies
In this paper, first we briefly review the during the pre-computer age found it less than
development of decision-support systems over the successful largely owing to the considerable
past two decades and the implications that these quantitative demands of the technique.16
have had for site forecasting. Secondly, we Wrigley regards the late 1980s as the ‘golden
examine how the role of the site visit has been age’ for store location analysis.17 Similarly,
assessed in the historical site selection literature Moutinho et al.18 identify the decade as a
and review its limited coverage in more landmark period for the sophistication of site
contemporary accounts. Thirdly, we look at two assessment procedures, characterised by the
case studies focusing on new site forecasting at a abandonment of ‘the intuitive approach to
major UK food retailer. By doing so we examine location decision-making’ (p. 205). They chart

140 Journal of Targeting, Measurement and Analysis for Marketing Vol. 16, 2, 139–155 © 2008 Palgrave Macmillan Ltd 0967-3237 $30.00
 The importance of context in store forecasting

Table 1: Principal site evaluation tools

Technique Details Technological
and data input

Experience/experimental ‘Rule of thumb’ procedures often employed ‘on site’ where the benefits of Low
experience, observation and intuition drive decision-making.


Checklist Procedure to systematically evaluate the value of (and between) site(s) on the
basis of a number of established variables.

Ratio Assumes that if a retailer has a given share of competing floorspace in an area it
will achieve a proportionate share of total available sales.

Analogues Existing store (or stores) similar to the site are compared to it to tailor turnover
expectations.

Multiple regression Attempts to define a correlation between store sales and variables within the
catchment that influence performance.

Geographical information Spatial representation of geodemographic and retail data that is based on
systems (GIS) digitalised cartography and draws on relational databases.

Spatial interaction Derived from Newtonian laws of physics based on the relationship between
modeling store attractiveness and distance from consumers. May operate ‘within’ a GIS.

Neural networks Computer-based models explicitly represent the neural and synaptic activity of High
the biological brain.

the development of publicly available geo- ‘the authors are not aware of any blue-chip
demographic databases in combination with businesses that are currently using this technology
increases in computing power and more technical for retail planning’ (Birkin, et al.,25 p. 144).
assessment techniques. This evolution was The clear thrust of recent academic literature
undeniably facilitated by the emergence of concerning store location assessment has been
geographical information systems (GIS) that ‘to tackle the uncertainties by developing more
were labelled in the early 1990s as a ‘paradigm sophisticated models in which they seek to
shift in cartography’.19 This spatial representation incorporate all variables and interactions’ (Dijst
of geo-demographic and retail data is particularly et al.,26 p. 1333). However, with many of these
powerful and based ‘on a crossing of digitalized progressive developments, it is easy to forget that
cartography in addition to relational databases’ retail location management is essentially a context-
(Mendes and Themido,20 p. 14) which ultimately specific practice where retail strategy, planning
allows non-GIS specialists to quickly interpret regulation, site availability and consumer behaviour
and understand complex geo-demographic intersect. Location planning is therefore concerned
patterns and trends.21 The development of GIS with a wealth of external factors that affect
effectively allowed the potential inherent in spatial proposals over and above the simple identification
interaction models to become a practical of the most efficient location for a store in
forecasting reality.22 relation to its catchment. For example, UK
As the technology to manipulate and analyse grocery retailers have had to become increasingly
huge amounts of data has developed, the potential flexible in location and format development, partly
of neural networks has been discussed in the driven by government planning legislation, leading
literature. This technique, while still in its infancy, to huge diversity in their store portfolios.27,28
moves into the domain of artificial intelligence by Indeed, Poole et al.29 underline the dynamism
explicitly representing the neural activity of the with this situation, noting that ‘the minimum
human brain.23 Despite the emphasis that this catchment population needed for store viability
method has received by academics,24 its application has declined over the last decade, meaning that
in practical forecasting has been more limited — more areas have potential for further exploitation

© 2008 Palgrave Macmillan Ltd 0967-3237 $30.00 Vol. 16, 2, 139–155 Journal of Targeting, Measurement and Analysis for Marketing 141
 Wood and Tasker

(especially by new retail formats)’ (pp. 2130–2131). advantages and impediments to success that it
Given such variation and innovation across a offers. Fenwick43 usefully distinguishes between
variety of locations, arguably ‘off the shelf ’ in-office locational advantages that ‘depend upon the
modelling packages may not provide the single characteristics of the surrounding population
‘solution’ to forecasting challenges; not least in and competition’ and site advantages that arise
the absence of numerous analogous stores within from its ‘peculiar characteristics’ and ‘range
the existing store estates of retailers which can from accessibility, store size, layout, proximity
act as benchmarks of performance when of competition etc’ (p. 3). Similarly, Brown44
forecasting sales. suggests that despite the recent advances in
location modelling and GIS, the outcome of
THE ROLE OF THE SITE VISIT locational decisions ultimately rests on micro-scale
IN STORE LOCATION considerations and the precise location within
DECISION-MAKING a catchment — exactly where the site visit is
Research that has analysed retailer’s employment of essential. Meanwhile, recent work in modelling
site assessment approaches has uncovered the store closures conceded that ‘we see that there
utilisation of a huge variation of techniques but are a number of location specific characteristics
has also noted that there are still many operators that may be of equal or greater importance
for whom ‘location planning is often undertaken than broader regional effects’ (Shields and
on the basis of subjective rules of thumb and a Kures,45 p. 267).
degree of opportunism relating to the availability When forecasting sales, analysts deal with issues
of individual sites’ (Pioch and Byrom,30 p. 223), that are not binary but relative and centre on
‘intuition’31 or ‘common sense’.32 Hence, the site issues of understanding and context as much as
assessment process is a blend of ‘art and science’,33 data processing. Indeed, recent research46 has
though the majority of the literature portrays it suggested consumers’ behaviour is becoming more
more as a complex office-based data manipulation difficult to forecast with changes in household
and modelling challenge. composition and responsibilities: ‘consumer choice
Even where complex quantitative modelling is is socially embedded within households’
employed, there is still a need to visit the store increasingly complex everyday lives, with
location and surrounding catchment in order to shopping “fitted in” around people’s other
assess the nature of that location and its place in responsibilities and commitments (childcare, work,
the context of wider competitive influences, leisure, etc)’ (p. 59). It is essential to note that
population, transport networks, as well as the accessibility is not simply the outcome of the
quality of unit or store site itself. From a geography of the catchment, but also ‘socially
practitioner standpoint, it is notable that the constructed notions of value, price, and quality’
academic literature continues to focus on largely (Jackson et al.,47 p. 61). It is the context-specific,
theoretical, context-unspecific and unapplied geographical, cultural and political nature of the
scenarios in technique development rather than catchment (and the interactions between them)
practical usage within the organisational context of the that has to be understood ‘on the ground’ as well
firm.34–39 Notably, Birkin et al.40 do not discuss as represented in data.48
the site visit, while Rogers’ practically focused Significantly for this paper, Wood and Browne49
1992 text41 makes only fleeting references to argue that there are considerably different site
this part of the process. For Clarke et al.42 there assessment challenges in the forecasting of smaller
is therefore a tendency for research to ‘either stores selling low order goods due to their limited
ignore or underplay retailers’ intuitive catchments that are inherently more difficult to
judgement in the location decision-making model. They find that the data upon which
process’ (p. 267). forecasting models for convenience stores are
In part, the importance of the site visit derives based is likely to be less rigorous at a finer spatial
from the micro-scale nature of the site and the scale of the micro environment where small

142 Journal of Targeting, Measurement and Analysis for Marketing Vol. 16, 2, 139–155 © 2008 Palgrave Macmillan Ltd 0967-3237 $30.00
 The importance of context in store forecasting

changes in store visibility or proximity to roads or aren’t present in a database of stores that already exist.
pedestrian flows may result in disproportionate (Wolfe,57 p. 9)
affects on sales performance. This places a greater Meanwhile, it is also clear that commercially
reliance on visiting the site in-person to available catchment data can have its limitations,
experience it from a consumer’s perspective rather which means calibrating via data collection at site
than rely predominantly on data manipulated in visits — an issue largely missing from recent store
the office via modelling. In contrast, larger stores location research:
have extensive catchments so are less sensitive to
the spatial scale of data and more likely to be While commercial geocoders become more accurate
accurately forecast using techniques such as the with every release, it is not uncommon to find
spatial interaction model. that — due to factors such as address errors — the
Given our argument that the essential role of geocoder returns a location that is 14 mile,12 mile
situational context is being largely overlooked in or even several miles from the true location.Yet,
many companies continue to make multi-million
the academic literature on store forecasting, what
dollar decisions from points entered into a geocoder
should the site visit actually entail and what can without any additional verification (Wolfe,58
it achieve? Table 2 presents a summary of this pp. 9–10)
literature, coupled with insights from the industry
experience of both authors. It details the role of
the site visit across three spatial scales of analysis, The challenge of knowledge
coupled with the sources of information for integration in site evaluation
assisting decision-making that it can provide. In Some management-based research has begun to
addition to analysis at various spatial scales, it is recognise the challenges of integrating qualitative
also essential to consider the time and day of the insights into the decision-making process. There
week the site visit is scheduled. Obviously have been some useful contributions reported in
viewing events such as the lunchtime trade in the site assessment literature, primarily from
town centres can provide a rather over-optimistic Clarke and colleagues in analysing the role of
representation of the potential in a catchment. intuition within store location decision-making.
The situational knowledge produced on the While this specific research does not consider the
site visit is well known within successful retailers site visit per se, it does use cognitive mapping to
to partially fill gaps in knowledge that can analyse differing conceptions regarding what
improve accuracy for forecasting. At a recent factors underpin successful store locations
analyst roundtable event organised by Estates between the different actors in the development
Gazette, Mark Chivers, Head of Strategy decision, whether site research specialists or senior
Development & Research at Boots, commented: management.59 Especially interesting was the
finding that senior managers in the decision-
It’s easy to sit in Nottingham and twiddle with the
GIS and think you can do a sales forecast for Hale
making process place a ‘greater reliance on key
in Cornwall.You go, and then you discover that the constructs and a higher proportion of non-factual
people there are different.56 information…benchmarking their thoughts
against their own experiences and outcomes of
Interestingly, within the industry, commentators previous corporate decisions’ (Clarke and
can be quite lucid about the limitations of in- Mackaness,60 p. 166). Latter research by the same
office modelling and therefore the requirement group61 devised a tool to select analogue stores by
for contextual data from the field: integrating the ‘soft’ insights of executives. While
Many factors that are not easily measurable (e.g. such analysis is undeniably interesting, it does not
operations) affect store performance, while other reveal how qualitative insights taken ‘in the field’
factors (e.g. visibility ratings) can only be measured can be integrated into modelling or justified in
in an imperfect manner. It is important to note that the decision-making process. Some older literature
retail models cannot directly model situations that has experimented with introducing intuition

© 2008 Palgrave Macmillan Ltd 0967-3237 $30.00 Vol. 16, 2, 139–155 Journal of Targeting, Measurement and Analysis for Marketing 143
 Wood and Tasker

Table 2: Role of the site visit

Catchment
Inventory of the competition Assess competition specifically for:

(a) Size of selling area

(b) No. of assistants
(c) No. of checkouts
(d) Range of goods and services
(e) Price policy
(f) Opening hours
(g) Additional services (eg petrol station, toilets, café)
(h) Car parking (no. of spaces and configuration)
(i) Condition of store (recently refitted?)
(j) Drivetimes to and from site in question
(k) Location of competitors (standalone/mall/district center/retail parks, etc)
(l) Nature of store performance (eg basket or trolley trade?)
(m) Observe core customers (eg age, affluence)

Study consumers through Customer ‘spotting’ surveys to understand:

surveys
(a) Current customer shopping patterns
(b) Perceptions of retail image of competitors and current stores in portfolio
(c) To study areas of under-penetration

Check residential areas Visit residential areas to review:

(a) Nature of residential catchment compared to available data (if any)
(b) Any areas of new housing development that may affect forecasts
(c) Cultural geography of the catchment. Understand divisions between areas that
may not be well represented in traditional data sets

Site location
Accessibility of the site and (a) Ease of access and egress in terms of to the site and within the site itself
throughout the catchment (eg car park layout)
(b) Role and perception of ‘trade barriers’ for the customers (eg rivers, motorways, topography,
etc)

Visibility of site (a) View from pedestrian walkways

(b) View from immediate road on entry and egress
(c) View from major adjacent roads

Traffic flows around site (a) Measure flows throughout different types of day
(b) Check road speeds and for one-way streets especially for model calibration
if using spatial interaction models
(c) Check for any new roads not recorded in current data or models

Pedestrian flows around Measure flows throughout different types of day

the site
Crime check Examine area around the site for evidence of crime, litter, etc

Site development scheme

Appraise the shape of the Appraise the scheme for:
store and car park
(a) Size and shape of store relative to the scheme plans
(b) Review the suitability of the car park shape and size relative to the scheme
(esp. in terms of access)
(c) Review the scheme critically –can it be improved?

Sources: Based on an extensive review of the literature (especially: Applebaum,50,51 Bowlby et al.,52 Cohen and Applebaum,53
Rogers54 and Snow and Scott55) as well as industry experience

within marketing forecasting models, concluding forecasting is understanding the flexibility and
that if managers and modellers could identify the variations required for different scenarios.
informational basis for exceptions to the model, Clearly then, ‘(m)any site factors are difficult to
the model refinement process could be improved.62 quantify but can have a dramatic effect on store
However, as our paper underlines, the art of site sales’ (Rogers,63 p. 207), but what improvements

144 Journal of Targeting, Measurement and Analysis for Marketing Vol. 16, 2, 139–155 © 2008 Palgrave Macmillan Ltd 0967-3237 $30.00
 The importance of context in store forecasting

can their consideration contribute to increasing about 350 spaces would have been the normal
accuracy of the forecasting process? Furthermore, requirement. In-office analysis of the potential
how can the knowledges that they produce be catchment showed that this was an underserved
integrated into the forecasting process? We now area in terms of supermarket provision and the
present two case studies based on store forecasting demographics very much matched the Sainsbury
at J Sainsbury plc, a major UK food retailer, that customer profile. Theoretically, this was a prime
seek to offer some perspectives on the opportunity but there were concerns regarding
contribution of the site visit and its integration the degree to which sales would be affected by
within the wider forecasting process. the constrained store size and parking. The site
analysts were of the opinion that the site must be
CASE STUDY 1: FORECASTING A visited to understanding the key context-specific
NEW SAINSBURY’S SUPERSTORE: issues before in-office modelling commenced.
SELSDON, CROYDON, UK
Site visit findings: The site itself
Prologue The visit highlighted the following issues about
In 2002, an opportunity became available to the site itself:
develop a Sainsbury’s supermarket in Selsdon,
Croydon, South London (see Figure 1). The site — The site was by a major crossroad and
was adjacent to Selsdon High Street and separated from the High Street by another,
comprised of a Local Authority car park, library less busy road. At peak time there was traffic
and playing fields. The extent of the area was very congestion at this junction with up to 50 cars
limited and would only accommodate a backing up in each direction.
supermarket with a 25,000 ft2 sales area and a — The store would be very prominent and visible
decked car park with just 245 spaces — given the from both roads and was already a destination
likely level of trade and access to the High Street, with the existing community centre and library.

M&S
CROYDON
ASDA

WALLINGTON

New Addington

SELSDON SITE

TESCO EXTRA

PURLEY 10 minute drivetime

WAITROSE
isochrone
Sanderstead

COULSDON
WAITROSE

Figure 1: Selsdon catchment with 10 min drivetime isochrone

© Crown Copyright Ordnance Survey. All rights reserved

© 2008 Palgrave Macmillan Ltd 0967-3237 $30.00 Vol. 16, 2, 139–155 Journal of Targeting, Measurement and Analysis for Marketing 145
 Wood and Tasker

— Just behind the site was a large independent trade strongly within Selsdon itself and where
school with many supermarket trips Selsdon stretched down to South Croydon
potentially linked with the ‘School run’. and Purley. In contrast to the output of the
— Although the High Street was the other gravity model, the analyst considered the
side of the Old Farleigh Road, there was a store would be unlikely to trade strongly
pedestrian crossing and bus stops nearby. from Forestdale and New Addington which
— The High Street was very lively with virtually benefited from being served by the new
no vacant units. Croydon Tramlink.
— At the other end of the High Street was a
Somerfield supermarket (8,000 ft2). Standards The competition
in the store were poor and the range was
focused on a top-up shop rather than a main — The only competition in Selsdon was
supermarket. The store was served by a small a Somerfield, which was effectively a
car park with 45 spaces. convenience store. Waitrose had consent
— The only other car parking was on-street pay to open an 18,000 ft2 store with just 100
and display and even off peak there were few surface spaces in Sanderstead but this would
vacant spaces. only serve the immediate community. More
significantly, in Purley there was a 59,000 ft2
Site visit findings: The catchment Tesco Extra with a 500-space surface car park
and petrol filling station.
— The catchment area showed a contrast of — Given its pre-eminence in the catchment, the
housing while Selsdon itself was an affluent Tesco store was visited on several occasions
outer London suburb with attractive mainly and was obviously very successful; clearly
semi-detached housing built in the 1930s. constrained by its size. However, the access
Most houses had their own garages and drives into and out of the store was very difficult
with more than one garage. due to a congested one-way road system.
— Located one mile to the south-west,
Sanderstead was similar but with larger more
exclusive houses and a higher proportion of Conclusions from the site visit and
detached houses. It was served by its own forecasting implications
small high street and there was a planning
consent for a small Waitrose. — The visit provided reassurance that the site
— Located one mile to the north-east, was prominent, visible and well placed to
Forestdale was a high density housing estate serve the identified catchment. The micro
built during the 1960s, consisting of privately environment was pleasant and no problems
owned dwellings especially suited to the first of security were evident.
time buyer. — The presence of a school, quality shops and
— Located two miles to the east, New bus routes meant the store was well placed
Addington was a post war council estate to pick up linked shopping trips.
displaying high degrees of private ownership — There was some concern about traffic
leading to significant redevelopment and congestion and access into and out of the
refurbishment. There was a strong parade car park. Strong recommendations were
of small shops, a community centre and made about installing a traffic light junction
swimming pool. with the right phasing. However, access
— It was noted from the visit that all of the issues would be no worse than the key
above were very distinct communities. competitor, Tesco.
Although within the 10 min driving time, the — A key requirement was identified by the analyst
visit indicated that the store should attract to re-balance some of the predicted draw of

146 Journal of Targeting, Measurement and Analysis for Marketing Vol. 16, 2, 139–155 © 2008 Palgrave Macmillan Ltd 0967-3237 $30.00
 The importance of context in store forecasting

trade within the catchment from the gravity to amend the modelled draw of trade. For a
model analysis subsequently undertaken in convenience store, the site visit actually forms the
the office. This store was going to achieve a basis of the estimate itself. At a practical level,
strong market share within Selsdon itself but arguably a gravity model is less appropriate in the
the propensity to draw from surrounding convenience store market due to the paucity of
communities would be challenging, with the available data designed for analysis at such a micro-
exception of linked school trips. This was scale.65 The trade is largely noncar and often linked
where the observations on the site visit led to with a separate shopping trip; leading to the
clearly identifiable changes in the forecasting catchment being analysed in a half mile radii rather
process. In particular, it is noteworthy that than a drivetime band. A regression model is the
this was based on no real ‘hard’ data but the most common tool to predict trade for such a unit
impression gained when visiting the site and was used in this instance in combination with
— similar to Clarke and Mackaness’ findings a site visit that was carried out first.
concerning the benchmarking of decisions Similar to forecasting supermarkets, the key
compared to previous analogous scenarios drivers of convenience store trade are population
previously observed.64 and competition, but to a lesser degree. However,
— The visit was reassuring in suggesting that it is not just number of households but also the
the constrained nature of the car parking was working population in the vicinity. ‘Workers’
probably going to be less of a factor given the population tends to exhibit a steeper distance
limited parking also offered by the competition. decay from the site that reflects the limited time
— The site visit gave reassurance that developing workers have before and after work as well as at
a store here was a viable decision and directly lunchtimes. Unfortunately, workplace data are
influenced the geography of the forecast rarely detailed enough to explain its micro-
catchment. geography in relation to the store location.
Competition in the convenience market is
The outcome extremely fragmented with a large number of
The store opened in June 2004 and traded just small multiple and independent operators that are
above expectation. Since the forecast, a Waitrose inadequately recorded in commercially available
opened as expected in Sanderstead with little databases. Furthermore, data on pedestrian foot-
impact on the stores sales as predicted following traffic are rarely available and when it does, is
the site visit. In December 2005, Tesco opened a unlikely to reveal the characteristics of the
15,000 ft2 mezzanine in its Purley store and have consumers passing the site; they could be
announced plans to redevelop the whole store to pensioners or city professionals. The importance
become the largest Tesco in the country. They of the site visit is also underlined in the need to
have also announced plans to build a store in understand the micro-location of the site: the
New Addington. The Somerfield in Selsdon has visibility and prominence of the site, the presence
closed and is being redeveloped as an Aldi. of on-street parking (often referred to as
‘stopability’) and bus stops are just some of the
factors that need to be established. The visit also
CASE STUDY 2: FORECASTING A determines whether the site has a ‘good pitch’ —
NEW SAINSBURY’S CONVENIENCE experience suggests that a few yards off the main
STORE IN HAYES, KENT retail area can reduce potential sales by half.

Background — Emphasising The site

contextual knowledge
The Selsdon case study clearly demonstrates the — The site was first identified in 1999 and was
impact a site visit had on the ultimate sales located on a local shopping parade in Hayes,
estimate for a supermarket in leading the analyst Kent, south-east England (see Figure 2).

© 2008 Palgrave Macmillan Ltd 0967-3237 $30.00 Vol. 16, 2, 139–155 Journal of Targeting, Measurement and Analysis for Marketing 147
 Wood and Tasker

WAITROSE
TESCO

CO-OP
CO-OP

MORRISONS

TESCO
EXPRESS HAYES
M&S
LOCAL
SIMPLY
FOOD
0.5 mile radii
ICELAND
CO-OP ICELAND EXTRA
NETTO

CO-OP
2 mile radii

Figure 2: Hayes local catchment with 0.5 and 2 mile radii

© Crown Copyright Ordnance Survey. All rights reserved

— Hayes shopping parade did not have a one- — There were no other major footfall drivers
stop shop supermarket, the nearest stores such as schools or major businesses found
being two Sainsbury’s superstores at a distance during the visit. Employee data indicated that
of 1.5 and 2.5 miles. The site was virtually only 1,000 people worked within ¼ mile of
opposite a small frozen food specialist retailer, the site, most of these being associated with
Iceland, towards the northern end of the local shops and services.
parade. At the other end of the parade about — Back at the office, the analysis of the
150 m away was Hayes Railway Station, a catchment demographics provided a mixed
southern terminus for a suburban line that runs picture: within one mile of the site there was
to Charing Cross in Central London. Opposite only a population of 20,000, although more
the station was a small convenience store. affluent than Greater London and GB as a
— The Sainsbury Local format requires a sales whole. From a positive perspective, pensioners
area of 3,000 ft2 and the site visit showed that accounted for about a third of the population
apart from Iceland and the site being reviewed, and there was an over-representation of
there were no other opportunities to establish a childless households. However, the proportion
store of this size in the immediate area. of single person households was below the
— There was on-street pay and display parking average for Greater London and GB. Also of
and a bus stop outside the site with a surface concern was the high car ownership (approx
car park behind the shopping parade opposite. 80 per cent), given the lack of good parking at
— The site was visited on a weekday and the site and the proximity of larger Sainsbury’s
Saturday morning and the footfall was typical stores on the edge of the localised catchment.
compared to other shops within the Sainsbury
portfolio and was predominately elderly. The model inputs
While the parade lacked some of the strong
major national retailers, all of the stores were — A simple regression model was used that was
occupied and the units were well maintained based on the Sainsbury Local portfolio of the
and appeared to be flourishing. time. Most of the inputs were score based,

148 Journal of Targeting, Measurement and Analysis for Marketing Vol. 16, 2, 139–155 © 2008 Palgrave Macmillan Ltd 0967-3237 $30.00
 The importance of context in store forecasting

which had a degree of subjectivity but was visits. Analysts are therefore encouraged to
backed by clear definitions for each score proactively amend in-store modelling when their
category. experience ‘on the ground’ gives them cause
— The population score was quite low reflecting to over-ride quantitative outputs.
the limited population and the adverse The store was re-merchandised to target the
characteristics. The working population top-up shopper and sales gradually increased over
also carried a low score reflecting both the following year. With an adjustment to the
the data and the experience of the visit. model to lower the transport node score, the
However, the site scored strongly in terms store ultimately traded on forecast, giving
of competition. Not only was there a lack reassurance that the model remained appropriate
of competition from the major operators and that it was the scoring process derived from
but also independent stores were also poorly the misinterpretation of the catchment on the site
represented. visit that had caused the inaccuracy.
— Footfall received an ‘average’ score reflecting
the counts undertaken during the two CONTEXTUAL KNOWLEDGE AND
visits; however, there was a high score for THE DECISION-MAKING PROCESS
prominence and visibility. ‘Stopability’ was The case studies presented in the above section
quite good with on-street parking and a serve as useful exemplars of the context-specific
surface car park 100 m away. nature of store decision-making and that
— Footfall drivers scored modestly with the forecasting cannot be reduced exclusively to office-
lack of strong adjacent operators, but the based processes and data management. In both
most difficult to score was ‘Transport nodes’. cases, analysts observed issues not represented in
There was a bus stop outside but the issue was their in-office data or models. In the Selsdon case,
whether there would be any sales the analyst was able to amend the modelled data
benefit from being 150 m from the rail station. following experience of the catchment. In the
While there was little evidence of second case, at Hayes, the analyst misinterpreted
commuters passing the site, it was felt that the impact of being ‘off-pitch’ and incorrectly
the presence of a Sainsbury’s Local would scored the transport variable within the regression
cause some consumers to divert to visit the model. However, in turn, lessons were learned and
store on their way home. This was a critical fed into the analysis of subsequent sites.
decision that stemmed from visit observations Although a well-developed knowledge base in
and had significant implications for forecast combination with complex modelling ability is
accuracy. a rich resource, its potential for solving location
forecasting issues can only be realised when it
is appropriately invoked and intelligently used.66
The outcome Importantly, while much of the management
The store opened in June 2000 and after a few literature is exploring issues of knowledge
months it became clear that it was trading at less management, situational context and decision-
than three quarters of its estimate. A review making processes,67,68 the literature on retail
indicated that it was failing to draw commuters location decision-making tends to focus on
down from the rail station. In effect, the store ‘harder’ and quantifiable knowledge, where
acted solely a ‘top-up shop’ rather than a ‘grab & known data are manipulated often at the
go’ store for commuter traffic. This provided a aggregate level, almost as if the decisions are
lesson for future forecasts regarding the penalty of purely a matter of technical judgment.69 Indeed,
locating slightly ‘off pitch’ when attempting to as Alexander70 observes, ‘we remain in danger of
capture commuter trade. Subsequent forecasts for accepting a partial understanding of the strategic
the Local format have therefore considered this as locational decision making process in which
a key issue to examine when carrying out site important cultural aspects are neglected’ (p. 62).

© 2008 Palgrave Macmillan Ltd 0967-3237 $30.00 Vol. 16, 2, 139–155 Journal of Targeting, Measurement and Analysis for Marketing 149
 Wood and Tasker

The conventional approach of business is to GIScience, yet it falls dramatically short of the kind of
logically analyse which inevitably leads to the rich and highly structured conceptions of space that
construction of ‘process’, or in this instance, are required to do justice to all the concerns of either
‘modelling procedures’, for employees to follow. the natural or social sciences. (p. 8)
Positively, this allows for replicability and best The very human agency that models may have
practice but: the effect of limiting may be the crucial piece
In practice this means: the more information, the of information that ensures an accurate forecast
better; ‘cool and calm’ strategic thinking should not (as with the Hayes case study).
be debased by feelings; efficient thought and behavior
must be called upon to subjugate emotion; and good Modelling ‘process’ and knowledge
organizations manage employees’ feelings out of the integration
process. (Sadler-Smith and Shefy,71 p. 77) There is a fundamental challenge in integrating
Our case studies have underlined though, that observations collected on the site visit into a
while the site visit generally provides data that form that can be integrated into technical
supports in-office analysis, it can also allow assessment techniques. To do so may require a
analysts to uncover issues that are sometimes change in the nature of that knowledge to be
difficult to represent in models (see Table 3). compatible with processing by conventional
At an elementary level the site visit may modelling techniques or ‘expert systems’. Clearly
provide more accurate data to enter into models some observations on the site visit are easier to
such as traffic flows or pedestrian footfall, thereby convert into codified knowledge than others. As
supporting the in-office analysis. In more Table 4 suggests, the collection of quantitative or
complex cases, the site visit exposes contextual codified data such as traffic flows are relatively
problems that cannot always be captured in easily integrated into models. In contrast, what we
objective functions and therefore solved as regard as ‘hybrid’ knowledge (in between codified
structured problems within models. As Longley72 and tacit knowledge), such as a field observation
argues, ‘digital representations of the real world of the visibility of a site or competitor store may
within GIS will almost always necessarily remain be interpreted in the field and some conclusions
partial and incomplete’ (p. 108). Moreover, for made by the analyst, but for this knowledge to
Fisher and Unwin73: become useful and included within the formalised
forecasting technique (eg a spatial interaction
the traditional mathematical conception of space model), this must be codified, possibly with
is widely perceived as constituting the dominant visibility graded on a numerical scale within
influence in providing a theoretical basis for computer code embedded within the model.

Table 3: Benefits of the site visit

Role of site Techniques to Calibration of spatial Collect data difficult to express Other benefits of the
visit employ ‘on site’ interaction models in spatial interaction models site visit

Examples Checklists Road speeds Visibility of site Aid in selection of suitable

of site visit analogue stores
benefit Experience/gut feel/ Competitor sizes Benefit/drawbacks of adjacent
experimental retailing
Competitor car park Role of ‘trade barriers’ (eg
(no of spaces) rivers, motorways) in customer
decision-making
Pedestrian flows ‘Quality’ of the competition
Traffic flows The ‘feel’ of the site
Population affluence Ease of access of store
versus database
Exact location of Type and structure of residential
competitors catchment

150 Journal of Targeting, Measurement and Analysis for Marketing Vol. 16, 2, 139–155 © 2008 Palgrave Macmillan Ltd 0967-3237 $30.00
 The importance of context in store forecasting

Meanwhile, pure tacit knowledge, intuition and Tuomi75 with such processes: ‘important aspects of
experience are rather more difficult to integrate knowledge are sedimented into the structure of
and may have to be considered ‘outside’ of the the measuring device’ (p. 109). The risk with such
formalised modelling process via some form of a distillation process and reconversion of field-
manual adjustment. based knowledge to binary data is that we may
The paradox with quantitative decision-support somehow lose the very understanding and insight
systems is often that ‘tacit knowledge has great that made the tacit knowledge gleaned in the
strategic value but that it must be codified to field so potentially beneficial. There is clearly a
exploit that value’ (Coff et al.,74 p. 454). For challenge to ‘off the shelf ’ forecasting packages
here to retain flexibility.
Table 4: Types of knowledge and knowledge integration in
the forecasting process The role of organisational context
Type of Example on site visit Difficulty to It is not just the ‘usability’ of the technique, but
knowledge integrate into also how it is implemented within the retailer
formalised
forecasting organisation and used in a practical context that is
process critical to success.76,77 There is a difference
Tacit ‘Feel’ of the catchment High between ‘knowledge’ per se (in theory) and
(eg quality of ‘knowing as action’78 as knowledge management
residential area, litter,
etc) also based on can only be understood in context, grounded in
‘experience’ what people do to get their work done.7
Hybrid Visibility of the site Medium (potential
The complexity of forecasting places huge
to introduce importance on individual analysts to maintain a
surrogate value strong understanding of the techniques that they
on a numerical
scale into model) are using and therefore the impact and limitations
of such data within the model. As part of this,
Codified Calibrate models with Low
observed numerical there must be the ability for the analyst to make
data (eg traffic flow, representations outside of the models if there are
store sales areas, etc).
issues above and beyond the model’s

Table 5: Knowledge conversion from the site visit and modelling interaction
Action Reducing observations to data and Considering knowledge outside of the model
incorporating into modelling systems

Site visit Observe, interpret and analyse the site

Knowledge generation Viewpoint formed from observation and

experience

Codification Reduce ‘knowledge’ to data to incorporate Do not incorporate into formal modelling process and
into model (eg gravity or regression model) take action to amend forecast outside of the modelling

Model Computational procedure

Output Output from model. Interpretation and

analysis

Forecast Determine numerical forecast

Determine cash return Determine profittability

on investment

Make decision Based on profittability but also broader

strategic perspectives

© 2008 Palgrave Macmillan Ltd 0967-3237 $30.00 Vol. 16, 2, 139–155 Journal of Targeting, Measurement and Analysis for Marketing 151
 Wood and Tasker

understanding and interpretation — as occurred strategy collide. Interpretative techniques to

in the Selsdon case study with manual understand this must be flexible and varied.
adjustments to the gravity model output and Furthermore, the lifecycle of a model must be
should have occurred in the Hayes example due considered in the decision-making process. The
to its slightly off pitch nature (see Table 5). As more complex the model, the greater the inputs,
such, ‘a combination of modelling and expert and ultimately the greater the work required to
judgement is an important way of improving keep those inputs up to date. This may mean that
decision effectiveness’ (Clarke et al.,79 p. 267), but a technique that performs initially well may end
clearly it remains challenging to develop ways of up impractical and resource draining in the longer
integrating these knowledges together in the term. Unfortunately, there is no single assessment
decision-making process. tool that provides the panacea to forecasting
difficulties. Instead it is often an untidy
CONCLUSIONS AND FUTURE combination of techniques — both from in-office
RESEARCH DIRECTIONS modelling and from site visits — that will reassure
There are some key findings that this study has the analyst that they are accurate in their
uncovered: first, and most simply, this paper estimates.
underlines the fact that despite huge advances in Secondly, it is essential that office-based models
the tools and assessment techniques of site do not become ‘black boxes’, impervious to all
appraisal, there is no substitute for the field visit but the most sophisticated statisticians and
and the observations across a range of spatial operational research (OR) specialists.81 Analysts
scales and times of day. While most observations are very active agents in the forecasting process
dovetail into in-office decision-support systems, and it is their understanding, interpretation and
they may be more difficult to introduce into analysis skills that are underplayed in the literature
formal modelling. However, these observations or to date. Indeed:
tacit knowledge can clearly contribute to more
accurate forecasting and thereby reduce Knowledge management systems are not automatic
investment risk. As the long established location data processing systems: knowledge management
expert, David Rogers80 recently reflected: initiatives therefore easily fail if they are conceived as
technology problems. The difficult thing, of course,
Technology cannot replace thorough field is that knowledge management then requires a broad
analysis and good retail intuition — nor cultural understanding of social, technical, and cognitive
understanding…. Too many site selection firms — on aspects of human organizations (emphasis added,
both sides of the Atlantic — mistakenly believe Tuomi,82 p. 115).
that the activity involves manipulating databases
and models in a comfortable office. While being a For example, with the case study in Selsdon,
great ‘assist’, location research technology is only as Croydon, the analyst was empowered to restrict
accurate as the data employed, and the judgments and the draw of trade forecast by the gravity model
care used to manage the process of application. (p. 64) following site visit observations, thereby leading
to a more accurate forecast. The role of the
Highly quantitative models are only analyst is very much one of fusing the ‘art and
simplifications of reality and rarely account for all science’ of location forecasting. Forecasters need
of the factors influencing a retail site, underlining to understand how their models work, the nature
the need to ‘get out there’. While it may be of their outputs and therefore how to amend
tempting for retail organisations to ‘replace’ site them in the light of situational and contextual
research executives with a technologically site visit data. In the implementation process of
advanced ‘model’, this is likely to prove a false new high-tech modelling solutions there is a key
economy. The location planning process is one role for the consultants of retailers to clearly
where issues of geography, consumer behaviour, explain to analysts how the models work —
government planning regulation and retailer hence the importance to building relationships

152 Journal of Targeting, Measurement and Analysis for Marketing Vol. 16, 2, 139–155 © 2008 Palgrave Macmillan Ltd 0967-3237 $30.00
 The importance of context in store forecasting

with such firms and ensuring good service is p. 1291). It is therefore essential to understand
critical.83 For instance, ‘there is no point in how these differing perspectives and narratives
developing an advanced sales forecasting method concerning knowledge management are played
that requires operation by a PhD in Statistics out.88 Contextual knowledge has to be
when the technique will, in fact, be administered sufficiently valued within the forecasting process
by market analysts with far less academic training’ even if it cannot be formally introduced within
(Rogers,84 p. 12). computational models that seem to provide
Third, a principal challenge raised by these ‘validity’ or ‘truth’. This, in turn, must also be
findings is resolving the conflict between codified considered, valued and reflected within the site
knowledge analysed and interpreted by assessment literature as it is only when
computational assessment techniques and systems modelling is employed in context can we
‘in the office’, and the tacit-based knowledges really evaluate its worth.
that are generated ‘in the field’. While the
strength of many of the latest highly technical Acknowledegments
techniques discussed is the highly formalised and Any comments made in this paper are the
structuring focus that they give to data responsibility of the authors and not necessarily
management, this can also make the introduction of the organisations discussed. We thank J
of the tacit knowledge collected ‘at site’ more Sainsbury plc for permission to use the case study
difficult to utilise. As such, ‘technologies are never material. We also appreciate the constructive
neutral: they embody values and make certain comments of Tony Hernandez, Andrew Alexander,
things possible, but not others’ (Tsoukas and Alan Hallsworth and two anonymous referees on
Mylonopoulos,85 p. S3). This challenge is earlier drafts of the paper. All errors or omissions
common across all organisations as it is essential are our own.
to ensure that knowledge management is
empowered rather than imposed by managerial References and Notes
and procedural approaches.86 1 Birkin, M., Clarke, G. and Clarke, M. (2002) ‘Retail Geography
and Intelligent Network Planning’, John Wiley, Chichester.
There is no reason why such flexibility cannot 2 Guy, G. (1999) ‘Exit strategies and sunk costs: The implications
be introduced into store forecasting processes for multiple retailers’, International Journal of Retail and
with the site visit and its observations forming a Distribution Management, Vol. 27, No. 6, pp. 237–245.
3 Cohen, S. and Applebaum, W. (1960) ‘Evaluating store sites
key part of the procedure, as occurred in the case and determining store rents’, Economic Geography, Vol. 36, No. 1,
studies. We need greater understanding of how pp. 1–35.
different kinds of knowledge are balanced and 4 Guy, G. (1999) op. cit.
5 González-Benito, J. and González-Benito, O. (2005) ‘The role of
mediated within forecasting and decision-making. geodemographic segmentation in retail location strategy’,
Given that decision-support systems are focused International Journal of Market Research, Vol. 47, No. 3, pp. 295–316.
on reducing risk and providing a return on 6 See Orr, J. (1996) ‘Talking about Machines: An Ethnography of
investment, the site visit is notable as requiring a Modern Job’, Cornell University Press, Ithaca, NY.
7 Orlikowski, W. (2002) ‘Knowing in practice: Enacting a
very little capital expenditure but has the collective capability in distributed organizing’, Organization
potential to provide considerable improvements Science, Vol. 13, No. 3, pp. 249–273.
in accuracy. 8 Barki, H. and Pinsonneault, A. (2005) ‘A model of organizational
integration, implementation effort, and performance’,
Clearly challenges are focused on the Organization Science, Vol. 16, No. 2, pp. 165–201.
organisational structure of retail and consultancy 9 Nonaka, I. (1994) ‘A dynamic theory of organizational knowledge
firms as much as the individual analysts creation’, Organization Science, Vol. 5, No. 1, pp. 14–37.
10 Nonaka, I. (1994) op. cit.
themselves and the techniques that they employ. 11 Hernandez, T., Bennison, D. and Cornelius, S. (1998) ‘The
Indeed, different expert communities within an organisational context of retail location decision making’,
organisation ‘may hold conflicting belief systems Geojournal, Vol. 45, No. 4, pp. 295–308.
and thus engender incongruent interpretations of 12 Jones, K. and Simmons, J. (1990) ‘The Retail Environment’,
Routledge, London.
a given item of knowledge, resulting in 13 Nelson, R. (1958) ‘The Selection of Retail Locations’,
communication breakdown’ (Hsiao et al.,87 FW Dodge Corp, New York.

© 2008 Palgrave Macmillan Ltd 0967-3237 $30.00 Vol. 16, 2, 139–155 Journal of Targeting, Measurement and Analysis for Marketing 153
 Wood and Tasker

14 Rogers, D. and Green, H. (1979) ‘A new perspective on 33 Ibid.

forecasting store sales: Applying statistical models and techniques 34 ReVelle, C. and Eiselt, H. (2005) ‘Location analysis: A synthesis
in the analog approach’, Geographical Review, Vol. 69, No. 4, and survey’, European Journal of Operational Research, Vol. 165,
pp. 449–458. pp. 1–19.
15 Davies, R. (1977) ‘Store location and store assessment research: 35 Dasci, A. and Laporte, G. (2005) ‘A continuous model for
The integration of some new and traditional techniques’, multistore competitive location’, Operations Research, Vol. 53,
Transactions of the Institute of British Geographers (New Series), No. 2, pp. 263–280.
Vol. 2, No. 2, pp. 141–157. 36 González-Benito, O. (2002) ‘Overcoming data limitations for
16 Breheny, M. (1988) ‘Practical methods of retail location analysis: store choice modeling. Exploiting retail chain choice data by
A review’, in Wrigley, N. (ed.), ‘Store Choice, Store Location means of aggregate logit models’, Journal of Retailing and
and Market Analysis’, Routledge, London, pp. 39–86. Consumer Services, Vol. 9, pp. 259–268.
17 Wrigley, N. (1988) ‘Retail restructuring and retail analysis’, in 37 Kuo, et al. (2002) op. cit.
Wrigley, N. (ed.), ‘Store Choice, Store Location and Market 38 Rust, R. and Donthu, N. (1995) ‘Capturing geographically
Analysis’, Routledge, London, pp. 3–34. localized misspecification error in retail store choice models’,
18 Moutinho, L., Curry, B. and Davies, F. (1993) ‘Comparative Journal of Marketing Research, Vol. 32, pp. 103–110.
computer approaches to multi-outlet retail site location 39 Sakashita, N. (2000) ‘An economic analysis of a convenience
decisions’, Service Industries Journal, Vol. 13, No. 4, pp. 201–220. store location’, Urban Studies, Vol. 37, No. 3, 471–479.
19 Morrison, J. L. (1994) ‘The paradigm shift in cartography: The 40 Birkin, et al. (2002) op. cit.
use of electronic technology, digital spatial data, and future 41 Rogers, D. S. (1992) ‘Retail Location Analysis in Practice’,
needs’, in Waugh, T. and Healey, R. (eds), ‘Advances in GIS Longman, Harlow.
Research’, Taylor and Francis, London, pp. 1–15. 42 Clarke, I., Horita, M. and Mackaness, W. (2000) ‘The spatial
20 Mendes, A. and Themido, I. (2004) ‘Multi-outlet retail site knowledge of retail decision makers: Capturing and interpreting
location assessment’, International Transactions in Operational group insight using a composite cognitive map’, International
Research, Vol. 11, pp. 1–18. Review of Retail, Distribution and Consumer Research, Vol. 10,
21 Hernandez, T. (2007) ‘Enhancing retail location decision support: No. 3, pp. 265–285.
The development and application of geovisualisation’, Journal of 43 Fenwick, I. (1978) ‘Techniques in store location research — A
Retailing and Consumer Services, Vol. 14, No. 4, pp. 249–258. review and applications’, Retail and Planning Associates
22 Birkin, et al. (2002) op. cit. Research Reports, PO Box 5, Corbridge, Northumberland,
23 Coates, D., Doherty, N., French, A. and Kirkup, M. (1995) NE45 5AA.
‘Neural networks for store performance forecasting: An 44 Brown, S. (1994) ‘Retail location at the micro-scale: Inventory
empirical comparison with regression techniques’, International and prospect’, Service Industries Journal, Vol. 14, No. 4,
Review of Retail, Distribution and Consumer Research, Vol. 5, No. 4, 542–576.
pp. 415–432. 45 Shields, M. and Kures, M. (2007) ‘Black out of the blue light:
24 Kuo, R., Chi, S. and Kao, S. (2002) ‘A decision support system An analysis of Kmart store closing decisions’, Journal of Retailing
for selecting convenience store location through integration of and Consumer Services, Vol. 14, No. 4, pp. 259–268.
fuzzy AHP and artificial neural network’, Computers in Industry, 46 Jackson, P., del Aguila, R. P., Clarke, I., Hallsworth, A., de
Vol. 47, pp. 199–214. Kervenoael, R. and Kirkup, M. (2006) ‘Retail restructuring and
25 Birkin, et al. (2002) op. cit. consumer choice 2. Understanding consumer choice at the
26 Dijst, M., Burrough, P. and Schot, P. (2005) ‘Coping with household level’, Environment and Planning A, Vol. 38, No. 1,
uncertainties in integrative spatial planning’, Environment and pp. 47–67.
Planning A, Vol. 37, pp. 1331–1334. 47 Ibid.
27 Guy, C. and Bennison, D. (2007) ‘Planning guidance and large- 48 See Miller, H. (2005) ‘What about people in geographic
store development in the United Kingdom: The search for information science?’, in Fisher, P and Unwin, D. (eds),
“flexibility”’, Environment and Planning A, Vol. 39, No. 4, ‘Re-Presenting GIS’, John Wiley, Chichester, pp. 215–242.
pp. 945–964. 49 Wood, S. and Browne, S. (2007) ‘Convenience store location
28 Wood, S., Lowe, M. and Wrigley, N. (2006) ‘Life after PPG6: planning and forecasting — A practical research agenda’,
Recent UK food retailer responses to planning regulation International Journal of Retail and Distribution Management, Vol. 35,
tightening’, The International Review of Retail, Distribution and No. 4, pp. 233–255.
Consumer Research, Vol. 16, No. 1, pp. 47–65. 50 Applebaum, W. (1965) ‘Measuring retail market penetration for a
29 Poole, R., Clarke, G. P. and Clarke, D. B. (2006) ‘Competition discount food supermarket — A case study’, Journal of Retailing,
and saturation in West European grocery retailing’, Environment Vol. 41, pp. 1–15.
and Planning A, Vol. 38, pp. 2129–2156. 51 Applebaum, W. (1966) ‘Methods for determining store trade
30 Pioch, E. and Byrom, J. (2004) ‘Small independent retail firms areas, market penetration, and potential sales’, Journal of Marketing
and locational decision-making: Outdoor leisure retailing by the Research, Vol. 3, pp. 127–141.
crags’, Journal of Small Business and Enterprise Development, Vol. 11, 52 Bowlby, S., Breheny, M. and Foot, D. (1985) ‘Store location:
No. 2, pp. 222–232. Problems and methods 4: Local site and store evaluation issues’,
31 Rogers, D. S. (1987) ‘Shop location analysis’, in McFadyen, E. Retail and Distribution Management, Vol. 2, March/April, pp. 40–44.
(ed.), ‘The Changing Face of British Retailing’, Newman, 53 Cohen and Applebaum (1960) op. cit.
London, pp. 74–83. 54 Rogers (1992) op. cit.
32 Hernandez, T. and Bennison, D. (2000) ‘The art and science of 55 Snow, W. and Scott, K. (1984) ‘Site appraisals’, in Davies, R. and
retail location decisions’, International Journal of Retail and Rogers, D. (eds), ‘Store Location and Store Assessment Research’,
Distribution Management, Vol. 28, No. 8, pp. 357–367. John Wiley, Chichester, pp. 215–231.

154 Journal of Targeting, Measurement and Analysis for Marketing Vol. 16, 2, 139–155 © 2008 Palgrave Macmillan Ltd 0967-3237 $30.00
 The importance of context in store forecasting

56 Tinworth, A. (2005) ‘Let’s find out where it’s at’, Estates Gazette, 73 Fisher, P. and Unwin, D. (2005) ‘Re-presenting geographical
Issue 540, 10th August, 2005. information systems’, in Fisher, P. and Unwin, D. (eds),
57 Wolfe, D. (2005) ‘10 common mistakes in site modeling’, ‘Re-presenting GIS’, John Wiley, Chichester, pp. 1–13.
MapInfo Magazine, Vol. 9, No. 1, pp. 9–11. 74 Coff , R., Coff, D. and Eastvold, R. (2006) ‘The knowledge-
58 Ibid. leveraging paradox: How to achieve scale without making
59 Clarke, et al. (2000) op. cit. knowledge imitable’, Academy of Management Review, Vol. 31,
60 Clarke, I. and Mackaness, W. (2001) ‘Management “intuition”: An No. 2, pp. 452–465.
interpretative account of structure and content of decision 75 Tuomi, I. (2000) ‘Data is more than knowledge: Implications of
schemas using cognitive maps’, Journal of Management Studies, the reversed knowledge hierarchy for knowledge management
Vol. 38, No. 2, pp. 147–172. and organizational memory’, Journal of Management Information
61 Clarke, I., Mackaness, W. and Ball, B. (2003) ‘Modelling intuition Systems, Vol. 16, No. 3, pp. 103–117.
in retail site assessment MIRSA: Making sense of retail location 76 Simkin, L. (1996) ‘Tackling barriers to effective implementation
using retailer’s intuitive judgments as support for decision- of modelling in retail marketing applications’, The International
making’, International Review of Retail, Distribution and Consumer Review of Retail, Distribution and Consumer Research, Vol. 6, No. 3,
Research, Vol. 13, No. 2, pp. 175–193. pp. 225–241.
62 Blattberg, R. and Hoch, S. (1990) ‘Database models and 77 Birks, D., Nasirin, S. and Zailani, S. (2003) ‘Factors influencing
managerial intuition: 50 per cent model +50 per cent manager’, GIS project implementation failure in the UK retailing industry’,
Management Science, Vol. 36, No. 8, pp. 887–899. International Journal of Information Management, Vol. 23,
63 Rogers, D. S. (2005) ‘Developing a location research pp. 73–82.
methodology’, Journal of Targeting and Analysis for Marketing, 78 Cook, S. and Brown, J. (1999) ‘Bridging epistemologies: The
Vol. 13, No. 3, pp. 201–208. generative dance between organizational knowledge and
64 Clarke, I. and Mackaness, W. (2001) op. cit. organizational knowing’, Organization Science, Vol. 10, No. 4,
65 Wood and Browne (2007) op. cit. pp. 381–400.
66 Armstrong, M., De, S., Densham, P., Lolonis, P., Rushton, G. and 79 Clarke, et al. (2000) op. cit.
Tewari, V. (1990) ‘A knowledge-based approach for supporting 80 Rogers, D. (2006) ‘Location research planning: The need for less
locational decisionmaking’, Environment and Planning B, Vol. 17, hype’, European Retail Digest, Vol. 49, pp. 63–64.
pp. 341–364. 81 Simkin (1996) op. cit.
67 Hsiao, R-L., Tsai, S. D-H. and Lee, C-F. (2006) ‘The problems 82 Tuomi (2000) op cit.
of embeddedness: Knowledge transfer, coordination and reuse 83 Kitay, J. and Wright, C. (2004) ‘Take the money and run?
in information systems’, Organization Studies, Vol. 27, No. 9, Organisational boundaries and consultants’ roles’, Service Industries
pp. 1289–1317. Journal, Vol. 24, No. 3, pp. 1–18.
68 Sadler-Smith, E. and Shefy, E. (2004) ‘The intuitive executive: 84 Rogers (1992) op. cit.
Understanding and applying “gut feel” in decision-making’, 85 Tsoukas, H. and Mylonopoulos, N. (2004) ‘Introduction:
Academy of Management Executive, Vol. 18, No. 4, pp. 76–91. Knowledge construction and creation in organizations’, British
69 Clarke, I. and Hallsworth, A. (1994) ‘Interorganizational networks Journal of Management, Vol. 15, pp. S1–S8.
and location investment decisions’, International Journal of Retail 86 Compare with Armistead, C. and Meakins,, M. (2002) ‘A
and Distribution Management, Vol. 22, No. 6, pp. 38–45. framework for practising knowledge management’, Long Range
70 Alexander, A. (1997) ‘Strategy and strategists: Evidence from an Planning, Vol. 35, pp. 49–71.
early retail revolution in Britain’, International Review of Retail, 87 Hsiao, et al. (2006) op. cit.
Distribution and Consumer Research, Vol. 7, No. 1, pp. 61–78. 88 Boland, R. and Tenkasi, R. (1995) ‘Perspective making and
71 Sadler-Smith, E. and Shefy, E. (2004) op. cit. perspective taking in communities of knowing’, Organization
72 Longley, P. (2004) ‘Geographical information systems: On Science, Vol. 6, No. 4, pp. 350–372.
modelling and representation’, Progress in Human Geography,
Vol. 28, No. 1, pp. 108–116.

© 2008 Palgrave Macmillan Ltd 0967-3237 $30.00 Vol. 16, 2, 139–155 Journal of Targeting, Measurement and Analysis for Marketing 155